Target Testing and Machine Learning Systems for Detecting and Identifying Machine Behavior

Abstract

Systems and methods for identifying the behavior of a machine are described. A computer system receives signals representing the operation of a live machine and applies a deep learning algorithm to identify patterns in a set of signals stored on a computer-readable memory. The set of signals includes the received signals representing the operation of the live machine and other signals. A series of target tests are performed using a test machine while monitoring signals representing the operation of the test machine. Behaviors that produce signals matching the patterns identified by the deep learning algorithm are identified during the series of target tests. Then, in response to detecting the pattern in the received signals representing the operation of the live machine, the behavior is automatically identified as occurring in the live machine.

Description

Field of the Invention

[0001] The present invention relates to systems and methods for detecting the condition or behavior of machines, such as heavy machinery and engineering vehicles. Summary of the Invention

[0002] In one embodiment, the present invention provides a method for identifying the behavior of a machine. A computer system receives signals representing the operation of a field machine and applies a deep learning algorithm to identify patterns in a set of signals stored on a computer-readable memory. The set of signals includes the received signals representing the operation of the field machine and other signals. A series of target tests are performed using a test machine while monitoring signals representing the operation of the test machine. During the series of target tests, a behavior that produces signals matching the patterns identified by the deep learning algorithm is identified. Then, in response to detecting the pattern in the received signals representing the operation of the field machine, the behavior is automatically identified as occurring in the field machine.

[0003] In another embodiment, the present invention provides a method for identifying the behavior of a machine by receiving multiple signals from multiple field machines. The multiple signals include the time-domain outputs of each of multiple sensors of each field machine. The multiple signals are stored in a computer-readable memory, and a deep learning algorithm is applied to the signals to identify multiple patterns in the signals. Then, a series of target tests are performed using a test machine while monitoring the time-domain outputs of each of the multiple sensors of the test machine. The series of target tests includes performing a series of operations under a set of defined varying operating conditions. During the series of target tests, when the outputs of the multiple sensors of the test machine match a first pattern of the multiple patterns identified by the deep learning algorithm, a first behavior is identified. A database defining multiple behaviors corresponding to different patterns of the multiple patterns is updated. Each behavior defined by the database includes the identified operation of the machine and the identified operating condition of the machine. Then, in response to detecting the first pattern in the signals from the multiple sensors of the field machine, the first behavior is automatically identified as occurring in one of the multiple field machines.

[0004] Other aspects of the present invention will become apparent by considering the detailed description and the drawings. Brief Description of the Drawings

[0005] Figure 1 is a block diagram of a local machine controller communicating with a remote server according to one embodiment.

[0006] Figure 2A flowchart of a method for identifying patterns in signal data from on-site vehicle sensors and identifying behaviors associated with the patterns through target testing.

[0007] Figure 3 A method for detecting emerging behaviors by detecting signal patterns previously identified and defined by Figure 2 A flowchart of a method.

[0008] Figure 4 A flowchart of a method for identifying signal data patterns / features using unsupervised machine learning, supervised machine learning, and target test data and associating them with machine operations.

[0009] Figure 5 A flowchart of a method for identifying signal data patterns / features using unsupervised machine learning, supervised machine learning, target test data, and manually defined classifications and associating them with machine operations.

[0010] Figure 6 A flowchart of a method for identifying signal data patterns / features, associating them with machine operations, and updating a supplementary target test plan. DETAILED DESCRIPTION

[0011] Before explaining any embodiments of the invention in detail, it is to be understood that the invention is not limited in its application to the details of construction and arrangement of components set forth in the following description or illustrated in the drawings. The invention is capable of other embodiments and of being practiced or carried out in various ways.

[0012] From several different starting points, it can be useful to identify the operating state and applications of a machine at a worksite. For example, knowing the operating state of a machine can be used to correlate authorizations with applications, to identify customer use of improved machine designs, and to identify patterns in machine use that can be used to develop (fully or partially) automated machines.

[0013] One mechanism for machine state recognition is to define machine states. These defined states (e.g., "idle", "transporting", "in cut", "swing to truck", "dump", "swing to trench", etc.) can be manually correlated with sensor readings that represent machine states (e.g., operator slew commands, pump pressure, engine speed, engine load, etc.). Then, a small set of physical tests can be used to determine the relevance or accuracy of the state machine. However, this is time-consuming, expensive, allows for a very narrow understanding of machine behavior, and only achieves a low level of correlation. Additionally, it can only be used to confirm an existing understanding of the machine because this approach requires knowledge of the features to be identified in the machine data before attempting to classify machine behavior. In some embodiments, types of machine learning - e.g., "deep learning", for example - can be combined with machine telematics to achieve a broader understanding of the operating state / application and corresponding sensor readings.

[0014] Figure 1 An example of a telematics system for a field machine 100 - e.g., a bulldozer, loader, track vehicle, or excavator, for example - communicating with a remote telematics server 120 is illustrated. In Figure 1 the example, the field machine 100 is equipped with a local machine controller 101 that includes an electronic processor 103 and a non-transitory computer-readable memory 105. The memory 105 stores instructions executed by the electronic processor 103 to provide, for example, the functionality described herein. The local machine controller 101 is communicatively coupled to one or more operator controls 107, and in some embodiments, the local machine controller provides output control signals to one or more machine actuators 109 to control the operation of the machine based (at least in part) on signals received from the operator controls 107. The local machine controller 101 is also communicatively coupled to one or more machine sensors 111, a display 113 (e.g., a user interface screen and / or indicator lights), and a transmitter and / or receiver 115.

[0015] In Figure 1In the example, the local machine controller 101 is configured to capture real-time data and transmit it to the remote telematics server 120. The real-time data includes, for example, inputs to the operator controls 107, sensor outputs from the machine sensors 111, and output control signals sent to the machine actuators 109. This data is transmitted from the transmitter / receiver 115 of the field machine 100 and received by the transmitter / receiver 121 of the remote telematics server 120. In some embodiments, the data from the field machine 100 is "real-time streamed" to the remote telematics server 120 nearly in real time, while in other embodiments, the data may be sent periodically or in larger compiled data files. Once received by the remote telematics server 120, the data is retained and stored in the database of the pattern detection server 123. In Figure 1 the example, the pattern detection server 123 includes an electronic processor 125 and a non-transitory computer-readable memory 127.

[0016] Although Figure 1 only a single field machine communicating with the remote telematics server 120 is shown, in some embodiments, a complete fleet of machines can be configured to stream real-time data back to the central database of the pattern detection server 123. In some embodiments, this database is configured to retain a complete history of every signal on the machine for each specific type of machine ever manufactured. It can represent the entire machine usage for that subset of machines. In some embodiments, the fleet usage data is then passed to a deep learning algorithm executed by the pattern detection server 123. The deep learning algorithm applies machine learning principles and sifts through the entire history of machine usage to identify patterns in the signals.

[0017] In some embodiments, the data is first analyzed to identify time-domain and / or frequency-domain patterns in the behavior of individual signals. A second analysis is then performed to identify correlations between these identified patterns. This provides insights into how the machine behaves, which patterns of behavior are relevant, and how the patterns might be related.

[0018] Deep learning algorithms provide lists of patterns, how the patterns are related, and how they can be recognized or defined. However, they do not provide any insights into what the patterns mean, what causes them, or how they can be related to machine operation or machine design. Instead, a similar machine is used to develop and execute a series of target tests while recording a full set of machine signals during the target relevance test. These signals are then processed to search for the patterns and correlations previously identified by the deep learning algorithm. By cross-referencing the target test plan and the actions performed during the target test with the resulting detected patterns and correlations, the physical meaning of the deep learning patterns is determined. In this way, the vehicle state can be understood in a more complex and emergent manner rather than through a prescriptive method.

[0019] Once the patterns and correlations are identified by the deep learning algorithm and their meaning is identified through the target tests, these patterns / correlations and thus the underlying behavior or system state can be identified and used in the field machine in real time, for example, to target how the machine is being used to optimize the performance of the machine for current use, to generate predictive diagnostic messages based on usage, to train the operator to perform better, or to automate part or all of the machine.

[0020] Figure 2 An example of a method for creating and / or updating a pattern / behavior database executed by a pattern detection server 121 is illustrated. First, the pattern detection server 121 receives one or more signals from the field machine (step 201). As discussed above, in some embodiments, these received signals can be combined with other signals from other machines in a central database. The pattern detection server 121 then applies a deep learning algorithm to the accumulated / received data to detect patterns and / or correlations of patterns (step 203). When a pattern and / or a correlation of a pattern is identified by the deep learning algorithm (step 205), the identification of the detected pattern is stored in a memory (step 207). For example, the system can be configured to store a full copy of the pattern itself, or can be configured to store one or more criteria for detecting / identifying the pattern or pattern correlation in the real-time sensor data of the field machine during operation.

[0021] After a pattern and / or a correlation of patterns is recognized by a deep learning algorithm, a series of target tests are developed and executed using a test machine (step 209). The test machine is a machine of the same type as the in-field machine, and the operation of the target tests is designed to be similar to the tasks performed by the in-field machine, which may generate the detected pattern or pattern correlation. The signals from the test machine are monitored as the test machine is used to execute the target tests (step 211), and the system (e.g., the pattern detection server or local controller of the test machine) detects when the signals of the test machine match a pattern or pattern correlation that has previously been recognized by the deep learning algorithm (step 213).

[0022] In response to detecting a pattern or pattern correlation in the test machine that matches a pattern or pattern correlation detected by the deep learning algorithm, it is learned that the target test being simultaneously executed is used to identify a particular behavior, condition, or system state corresponding to the particular pattern or pattern correlation (automatically or manually by the controller / server). The pattern / behavior database is then updated to include the pattern / pattern correlation recognized by the deep learning algorithm and the behavior identified by the target test as corresponding to the recognized pattern / correlation (step 215). This process is repeated when other signals are received from the in-field machine, when other patterns / correlations are recognized by the deep learning algorithm, and when other associated behaviors are recognized by the target test, and the pattern / behavior database is repeatedly updated. The updated pattern / behavior database is then sent to the in-field machine for use during the operation of the in-field machine (step 217).

[0023] As discussed above, the "target test" includes a series of tests designed to be performed by a test machine (which is the same type of machine as the field machine) and includes tasks similar to those to be performed by the field machine, which tasks can generate the detected patterns, pattern correlations, or other features in the collected field data. A specific example of a target test that can be performed includes an excavator trenching, where the excavated soil is deposited in a spoil pile parallel to the trench. A series of specific tests in this test plan can include, for example: (a) lowering the boom towards the ground, (b) starting to dig or "cut" with the bucket, (c) lifting the soil; (d) turning the boom towards the spoil pile while transporting the soil in the bucket, (e) releasing the soil into the spoil pile, and (f) turning the boom back towards the trench with no soil in the bucket. Other examples of target tests may include: excavator trenching where the excavated soil is deposited in a truck, where the truck tires are flush with the excavator tracks; or large-scale excavation where the excavator is on an elevated work platform and the excavated soil is deposited in a truck, where the truck bed is flush with the excavator tracks. The "test plan" can be developed using one or more of these specific examples described above, or can include other operations or operating environments instead of or in addition to the examples discussed above. Test plans can also be developed for other field machines and / or other specific uses / operations.

[0024] Moreover, as discussed further below in detail, a "test plan" can be developed to include a series of "test tags" corresponding to specific operations performed by the machine during the target test (e.g., lifting the soil). During the execution of the target test using the test machine, machine data is collected and associated with specific "test tags", and then in some embodiments the "test tags" are used as inputs to "supervised machine learning" operations.

[0025] Figure 3 An example of a method performed by a field machine to utilize a pattern / behavior database during real-time operation of the field machine is shown. After the field machine receives an updated pattern / behavior database (step 301), the field machine is operated by the user (step 303), and the signals generated by the field machine are monitored (step 305). As discussed above, these signals are either streamed in real-time to a remote telematics server or stored for later transmission to a pattern detection server (step 307).

[0026] When the local machine controller 101 of the on-site machine detects a pattern or pattern correlation stored in the pattern / behavior database (step 309), it also identifies the behavior defined by the pattern / behavior database corresponding to the detected pattern or pattern correlation (step 311). In some embodiments, the local machine controller may be configured to perform a specific action in response to detecting a specific behavior or condition. In other embodiments, the pattern / behavior database may further be configured to define a specific action to be performed by the local machine controller in response to detecting a specific pattern or pattern correlation. In Figure 3 the example of, the local machine controller is configured to determine whether an action or mitigation is defined by the database corresponding to the pattern or pattern correlation that has been identified in the signals of the on-site machine (step 313). If the pattern / behavior database defines the corresponding action or mitigation, the local machine controller will impose the action or mitigation in response to detecting the specific pattern or pattern correlation (step 315).

[0027] In some embodiments, the system may be configured to maintain and report to a remote server a record of how many times a specific behavior has been detected and how long each occurrence lasted. In other embodiments, maintaining this type of record may be an action defined by the pattern / behavior database for a specific behavior / signal. Other examples of actions that may be defined by the pattern / behavior database to be performed in response to detecting a specific behavior / signal may include adjusting the operating parameters of the machine or outputting an alert indicating that a specific behavior has been detected to the operator (e.g., via display 113).

[0028] A specific example of an action that may be initiated based on one or more detected occurrences of a behavior / signal may include cross-referencing machine patterns for necessary repairs. Finding the correlation of one or more fleet-wide behaviors that tend to lead to a certain type of repair can be used to predict which machines will require repairs, identify abusive behaviors, enable operators to be trained not to perform abusive behaviors, enable engineering personnel to redesign components, or perform a more accurate root cause analysis for repairs that were not anticipated during design.

[0029] These patterns and behaviors can also be used (e.g., by a manufacturer or machine dealer) to determine when a particular user / customer has the wrong machine size or type for a particular usage profile. By comparing behaviors on machines of different sizes, the pattern detection server can be configured to identify patterns associated with breakdowns on one machine type / size but not another. This can mean that if a customer / user purchases a larger machine, they can avoid damaging their machine and the need for repairs. This can also be used to identify those using an oversized machine. For customers who own / operate a fleet of machines, this can provide valuable data in terms of which job to use which of their machines, in order to optimize their fleet usage.

[0030] Another potential action that can be taken in response to detecting one or more occurrences of a particular pattern or pattern correlation is to verify machine automation features and automated machines for in-field use of the machine. If the automation features generate the same usage patterns as an expert operator, it can be confirmed that the automation features are performing with the same quality and efficiency. This can be used as a verification of the implemented automation features, or as part of supervised or reinforcement training during the development of automated features. In supervised learning, the control agent varies how it applies machine commands in an attempt to more closely reproduce the patterns seen in an expert operator. In reinforcement learning, creating patterns is one of many metrics that can be used as a reward to help a machine learn to optimize its behavior.

[0031] The examples presented above merely illustrate several examples of how the system can be configured to use "target testing" to identify and classify machine behavior. Other specific implementations are possible. For example, in some implementations, a two-stage "machine learning" approach is utilized that includes "unsupervised learning" (i.e., identifying patterns and / or features in the collected field data) performed on the collected field data and "supervised learning" (i.e., correlating the classified patterns / features with specific operations of the machine) performed on the machine data collected during target testing. Figures 4 to 6 Some detailed examples including such a two-stage machine learning approach are presented.

[0032] In the Figure 4 method, machine data is collected from multiple different in-field machines in a fleet (step 401). Unsupervised machine learning is performed on the collected fleet machine data (step 403) to generate a UML ("unsupervised machine learning") feature classifier (step 405). The UML feature classifier is configured to receive time-series machine data and classify portions of the received data according to a set of classifications previously identified / determined from the collected fleet data. However, at this stage, the UML feature classifier is not able to determine what operation (if any) the classified features might represent.

[0033] A target test plan is also generated (step 407), and the target test plan is executed using a test machine (step 409). Machine data is collected from the test machine during the target test and corresponding "test labels" are output from the target test (steps 411 and 413). The machine test data is then provided as input to a UML feature classifier (step 415), and the machine test data is classified based on features identified / detected during unsupervised machine learning. The classification of the test data is then passed to a supervised deep learning algorithm, which correlates the classification with the test label (step 417). Supervised machine learning determines which combinations of the features determined by the UML feature classifier are characteristic of each of the tests executed during the target test (step 419). In some embodiments, the supervised machine learning performed on the test data results in a classifier that is configured to receive machine data from in-field machines in real time and output a representation of the particular operation being performed by the in-field machines.

[0034] Figure 5 Another method that utilizes both supervised and unsupervised machine learning is illustrated, but the method is further enhanced by other manual or predefined classifications. Like Figure 4 the example of, unsupervised machine learning (step 501) is performed on the fleet machine data (step 503) to generate a UML feature classifier (step 505). A target test plan is generated (step 507), and the target test plan is executed using a test machine (step 509) to produce a set of machine test data (step 511) and test labels (step 513). However, in Figure 5 the example of, in addition to the UML feature classifier (step 517), additional manual (or other predefined classification criteria) are applied to the machine test data (step 515). The output of the manual classification, the output of the UML feature classifier, and the test labels are then provided as input to supervised machine learning (step 519), and supervised machine learning determines which combinations of the features determined by the UML feature classifier are characteristic of each of the tests executed during the target test (step 521).

[0035] For example, a target test plan can include an excavator cycle consisting of operations that occur in the following order: CUT operation, LOADED REPOSITION operation, DUMP operation, and UNLOADED REPOSITION operation (starting with the UNLOADED REPOSITION operation). Then, a UML feature classification algorithm can be applied to the test data, and then transitions can be made between these operations based on manually determined criteria. For example, when a swing command is initiated during the CUT operation, the transition from the CUT operation to the LOADED REPOSITION operation can be represented by the manually determined criteria. Similarly, when a "bucket dump" command is initiated during the LOADED REPOSITION operation, the transition from LOADED REPOSITION to DUMP can be represented by the manually determined criteria. When a swing command in the opposite direction of the LOADED REPOSITION operation is initiated during the DUMP operation, the transition from DUMP to UNLOADED REPOSITION can be represented by the manually determined criteria. Finally, during the UNLOADED REPOSITION operation, when the measured pump pressure (e.g., for one or more hydraulic cylinders of the excavator arm) becomes high (e.g., above a threshold), the transition from UNLOADED REPOSITION to CUT can be represented by the manually determined criteria.

[0036] Finally, Figure 6 Illustrated is a method in which a target test plan is supplemented / updated and additional tests are performed until all features identified in the fleet machine data are related to specific operations performed during the target test. Again, unsupervised machine learning (step 601) is performed on the fleet machine data (step 603) to generate a UML feature classifier (step 605). A target test plan is generated (step 607), and the target test plan is executed using a test machine (step 609) to produce a set of machine test data (step 611) and test labels (step 613). The UML feature classifier is applied to the machine test data collected during the target test (step 615) to identify the UML feature classifications in the collected test data, and supervised machine learning is performed to relate the features classified from the test machine data to specific test labels (step 617). The supervised machine learning determines which combinations of the features determined by the UML feature classifier are characteristic of each of the tests performed during the target test (step 619).

[0037] However, after performing supervised machine learning, the system examines a set of UML features identified from the fleet machine data by unsupervised machine learning (step 621) and determines whether all those identified features are also present / detected in the test data (step 623). If not, the test data does not cover the classification and additional testing is necessary to represent all the behaviors seen in the fleet data. Accordingly, complementary tests are identified and added to the test plan (step 625), and target testing continues according to the complementary test plan (step 609) until all behaviors have been identified from the fleet machine data in at least one test.

[0038] Accordingly, the present invention particularly provides a system and method for using deep learning to identify patterns or pattern correlations in machine signals, for attributing behaviors associated with the patterns by target testing, and for operating field machines based on the identified signal patterns and the defined corresponding behaviors. Various features and advantages of the present invention are set forth in the claims.

Claims

1. A method for identifying the behavior of a machine, the method comprising: Receiving, by a computer system, a signal representing the operation of a field machine; Applying, by the computer system, a deep learning algorithm to the received signal representing the operation of the field machine, wherein the deep learning algorithm is configured to discover new patterns in a set of signals stored on a computer-readable memory, wherein, for any particular behavior corresponding to the field machine, the new patterns are unknown, and the set of signals includes the received signal representing the operation of the field machine; After discovering the new patterns in the set of signals, operating a test machine to perform a series of target tests; While performing the series of target tests, detecting the occurrence of the new patterns in the signals representing the operation of the test machine; Based on the target tests performed by the test machine when the occurrence of the new patterns is detected in the signals representing the operation of the test machine during the series of target tests, identifying, during the series of target tests, the behavior of the test machine that causes the occurrence of the new patterns to be detected in the signals representing the operation of the test machine; Associating the identified behavior with the new patterns; and In response to a subsequent occurrence of the new patterns being detected in the received signal representing the operation of the field machine, automatically detecting a subsequent occurrence of the identified behavior in the field machine.

2. The method according to claim 1, wherein, Using the test machine to perform the series of target tests includes: performing a series of prescribed operations under a set of defined varying operating conditions, wherein the series of prescribed operations and the set of defined varying operating conditions are designed to reconstruct the new patterns in the signals representing the operation of the test machine.

3. The method according to claim 1, wherein Receiving a signal representing the operation of the field machine includes: receiving a signal representing the output of at least one sensor of the field machine.

4. The method according to claim 1, wherein, Applying the deep learning algorithm includes: discovering in the signals representing the operation of the field machine at least one pattern selected from the group consisting of time domain patterns and frequency domain patterns.

5. The method according to claim 1, further comprising: Receiving, by the computer system, a plurality of signals from a plurality of field machines, the plurality of signals including the time domain outputs of each of a plurality of sensors of each of the plurality of field machines, wherein receiving the plurality of signals from the plurality of field machines includes receiving the signal representing the operation of the field machine.

6. The method according to claim 1, wherein, Receiving a signal representing the operation of the field machine includes: receiving a signal representing at least one selected from the group consisting of an operator rotation command signal, a pump pressure signal, an engine speed signal, and an engine load signal.

7. The method according to claim 1, further comprising: Updating, by the computer system, a database that identifies a plurality of behaviors, each of the plurality of behaviors corresponding to a different pattern among a plurality of patterns, wherein each of the plurality of behaviors includes an identified operation of the machine and an identified operating condition of the machine.

8. The method according to claim 1, further comprising: Based on the received signal representing the operation of the field machine, store a record of each occurrence of the detected behavior for the field machine in the computer-readable memory.

9. The method according to claim 1, further comprising: In response to detecting a subsequent occurrence of the new pattern in the received signal from the field machine, send an operation adjustment signal from the computer system to the field machine, wherein the field machine is configured to adjust its operation in response to receiving the operation adjustment signal.

10. The method according to claim 1, wherein, The identified behavior includes: operating a machine that requires specific repair or maintenance, and The method further includes: in response to detecting the occurrence of the new pattern in the received signal representing the operation of the field machine, sending a signal identifying the specific repair or maintenance required for the field machine from the computer system.

11. The method according to claim 1, wherein, The identified behavior includes: using the field machine to perform a task that is not the best fit for the field machine, and The method further includes: in response to detecting the occurrence of the new pattern in the received signal representing the operation of the field machine, automatically sending a signal from the computer system advising the end user to use a different type of field machine for a specific task.

12. A method of identifying the behavior of a machine, the method comprising: Receiving, by a computer system, a plurality of signals from a plurality of field machines, the plurality of signals including the time-domain output of each of a plurality of sensors of each of the plurality of field machines; Storing, by the computer system, the plurality of signals in a computer-readable memory; Applying, by the computer system, a deep learning algorithm to the plurality of signals, wherein the deep learning algorithm is configured to discover a plurality of patterns in the plurality of signals stored in the computer-readable memory, wherein the patterns discovered are unknown for any specific behavior corresponding to the field machine, and wherein the plurality of patterns discovered by the deep learning algorithm includes a first pattern discovered in the signal received from a first sensor of at least one of the plurality of field machines; After discovering the first pattern, operating a test machine to perform a series of target tests, wherein the series of target tests includes performing a series of operations under a set of defined varying operating conditions; While performing the series of target tests, detecting the occurrence of the first pattern in the signal received from the first sensor of the test machine; During the series of target tests, identifying a first behavior of the test machine, wherein the identified first behavior occurs simultaneously with the detected occurrence of the first pattern in the signal received from the first sensor of the test machine; The database is updated by the computer system, which identifies a plurality of behaviors, each of the plurality of behaviors corresponding to a different pattern among the plurality of patterns discovered by the deep learning algorithm, wherein each of the plurality of behaviors includes an identified operation of the machine and an identified operating condition of the machine, and wherein updating the database includes updating the database to associate the first behavior with the first pattern; and In response to detecting a subsequent occurrence of the first pattern in a signal received from the first sensor of one of the plurality of field machines, the computer system automatically detects a subsequent occurrence of the first behavior in the one of the plurality of field machines.

13. The method according to claim 12, wherein, Receiving the plurality of signals from the plurality of field machines includes: receiving a periodic output or a continuous output from a transmitter of a first field machine among the plurality of field machines.

14. The method according to claim 12, wherein, Receiving the plurality of signals from the plurality of field machines includes: receiving at least one signal selected from the group consisting of an operator rotation command signal, a pump pressure signal, an engine speed signal, and an engine load signal.

15. The method according to claim 12, wherein, Applying the deep learning algorithm includes: discovering at least one pattern selected from the group consisting of a time domain pattern and a frequency domain pattern.

16. The method according to claim 12, further comprising: Discovering a pattern combination including a second pattern and a third pattern identified by the deep learning algorithm, wherein occurrences of the first pattern and the second pattern are discovered by the deep learning algorithm in time-corresponding outputs from two different sensors in a first field machine, During execution of the series of target tests, detecting an occurrence of the pattern combination in output signals from two different sensors of a test machine; Identifying a second behavior of the test machine during the series of target tests, wherein the identified second behavior occurs simultaneously with the occurrence of the pattern combination; Updating the database to associate the second behavior with the pattern combination; and In response to detecting a subsequent occurrence of the pattern combination in signals received from the two different sensors of one of the plurality of field machines, automatically detecting a subsequent occurrence of the second behavior in the one of the plurality of field machines.

17. The method according to claim 12, further comprising: Storing a record of each occurrence of the first behavior detected for each individual field machine among the plurality of field machines in the computer-readable memory.

18. The method according to claim 17, wherein, The database represents that the first pattern corresponds to a field machine being used for a task, the field machine not being the best-suited for the task, and The method further comprises: automatically outputting a report identifying a different type of field machine that is more suitable for the end user based on the number of occurrences of the first behavior detected for the first field machine.

19. The method according to claim 12, wherein, The database represents that the first pattern corresponds to a field machine being used for a task, the field machine not being the best-suited for the task, and The method further includes: automatically sending from the computer system a signal advising an end user to use a different type of field machine for the task.

20. The method according to claim 12, further comprising: In response to detecting the first pattern in the received signal from the first field machine among the plurality of field machines, sending an operation adjustment signal from the computer system to the first field machine, wherein the first field machine is configured to adjust its operation in response to receiving the operation adjustment signal.

21. The method according to claim 12, wherein, The database further identifies a specific repair or maintenance task corresponding to the first pattern, and the method further includes: In response to detecting the first pattern in the received signal from the first field machine among the plurality of field machines, sending from the computer system a signal indicating that the specific repair or maintenance is required for the first field machine.

22. The method according to claim 21, further comprising: In response to receiving the signal indicating that the specific repair or maintenance is required from the computer system, automatically scheduling the specific repair or maintenance for the first field machine.

Images